Search results

test-vsie-softmax host compile fails: vsie-softmax.cuh pulls in CUDA-only vsie-ptx.cuh

Status: Open.

#1706 In ikawrakow/ik_llama.cpp;

· wahargis opened on Apr 28, 2026

PP performance differences and config flags for large MoE model (Qwen3.5-397B) (slower than expected)

Status: Open.

#1695 In ikawrakow/ik_llama.cpp;

· chrisoutwright opened on Apr 25, 2026

LFM2.5 Support

Status: Open.

#1694 In ikawrakow/ik_llama.cpp;

· the7imperial7state-cmd opened on Apr 25, 2026

Gemma-4 MoE IQ4_XS: NaN logits → sampling crash / CUDA illegal memory access after #1657

Status: Open.

#1693 In ikawrakow/ik_llama.cpp;

· Geekmemz opened on Apr 25, 2026

Bug: Qwen 3.5 397B crashes on large prompts on V100

Status: Open.

#1692 In ikawrakow/ik_llama.cpp;

· jkyamog opened on Apr 25, 2026

Bug: Last checkpoint before TG not saved when images are in context (Qwen 3.5)

Status: Open.

#1676 In ikawrakow/ik_llama.cpp;

· MrHills-rs opened on Apr 22, 2026

Community Resource: Python bindings for ik_llama.cpp (ik-llama-cpp-python)

Status: Open.

#1674 In ikawrakow/ik_llama.cpp;

· gongpx20069 opened on Apr 22, 2026

Feature Request: Backend optimization for speculative decoding

enhancementNew feature or requestNew feature or request

Status: Open.

#1630 In ikawrakow/ik_llama.cpp;

· SamuelOliveirads opened on Apr 13, 2026

Feature Request: merge_up_gate_shexp

enhancementNew feature or requestNew feature or request

Status: Open.

#1629 In ikawrakow/ik_llama.cpp;

· Nexesenex opened on Apr 13, 2026

MoE models are broken with CUDA + CPU and no layers offloaded to the GPU

Status: Open.

#1618 In ikawrakow/ik_llama.cpp;

· ZenSirzechs opened on Apr 11, 2026

Bug: Segfault on GLM 5.1 with spec-decoding

Status: Open.

#1612 In ikawrakow/ik_llama.cpp;

· ValeryNo opened on Apr 10, 2026

Bug: Missing ik-llama-builder

Status: Open.

#1608 In ikawrakow/ik_llama.cpp;

· DevilaN opened on Apr 10, 2026